]>
Commit | Line | Data |
---|---|---|
9e853f23 RZ |
1 | /* |
2 | * Persistent Memory Driver | |
3 | * | |
9f53f9fa | 4 | * Copyright (c) 2014-2015, Intel Corporation. |
9e853f23 RZ |
5 | * Copyright (c) 2015, Christoph Hellwig <hch@lst.de>. |
6 | * Copyright (c) 2015, Boaz Harrosh <boaz@plexistor.com>. | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or modify it | |
9 | * under the terms and conditions of the GNU General Public License, | |
10 | * version 2, as published by the Free Software Foundation. | |
11 | * | |
12 | * This program is distributed in the hope it will be useful, but WITHOUT | |
13 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
14 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
15 | * more details. | |
16 | */ | |
17 | ||
18 | #include <asm/cacheflush.h> | |
19 | #include <linux/blkdev.h> | |
20 | #include <linux/hdreg.h> | |
21 | #include <linux/init.h> | |
22 | #include <linux/platform_device.h> | |
23 | #include <linux/module.h> | |
24 | #include <linux/moduleparam.h> | |
b95f5f43 | 25 | #include <linux/badblocks.h> |
9476df7d | 26 | #include <linux/memremap.h> |
32ab0a3f | 27 | #include <linux/vmalloc.h> |
34c0fd54 | 28 | #include <linux/pfn_t.h> |
9e853f23 | 29 | #include <linux/slab.h> |
61031952 | 30 | #include <linux/pmem.h> |
9f53f9fa | 31 | #include <linux/nd.h> |
32ab0a3f | 32 | #include "pfn.h" |
9f53f9fa | 33 | #include "nd.h" |
9e853f23 RZ |
34 | |
35 | struct pmem_device { | |
9e853f23 RZ |
36 | /* One contiguous memory region per device */ |
37 | phys_addr_t phys_addr; | |
32ab0a3f DW |
38 | /* when non-zero this device is hosting a 'pfn' instance */ |
39 | phys_addr_t data_offset; | |
c4544205 | 40 | u64 pfn_flags; |
61031952 | 41 | void __pmem *virt_addr; |
cfe30b87 | 42 | /* immutable base size of the namespace */ |
9e853f23 | 43 | size_t size; |
cfe30b87 DW |
44 | /* trim size when namespace capacity has been section aligned */ |
45 | u32 pfn_pad; | |
b95f5f43 | 46 | struct badblocks bb; |
9e853f23 RZ |
47 | }; |
48 | ||
59e64739 DW |
49 | static void pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset, |
50 | unsigned int len) | |
51 | { | |
5a92289f | 52 | struct device *dev = pmem->bb.dev; |
59e64739 DW |
53 | sector_t sector; |
54 | long cleared; | |
55 | ||
56 | sector = (offset - pmem->data_offset) / 512; | |
57 | cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len); | |
58 | ||
59 | if (cleared > 0 && cleared / 512) { | |
60 | dev_dbg(dev, "%s: %llx clear %ld sector%s\n", | |
61 | __func__, (unsigned long long) sector, | |
62 | cleared / 512, cleared / 512 > 1 ? "s" : ""); | |
63 | badblocks_clear(&pmem->bb, sector, cleared / 512); | |
64 | } | |
65 | invalidate_pmem(pmem->virt_addr + offset, len); | |
66 | } | |
67 | ||
e10624f8 | 68 | static int pmem_do_bvec(struct pmem_device *pmem, struct page *page, |
9e853f23 RZ |
69 | unsigned int len, unsigned int off, int rw, |
70 | sector_t sector) | |
71 | { | |
b5ebc8ec | 72 | int rc = 0; |
59e64739 | 73 | bool bad_pmem = false; |
9e853f23 | 74 | void *mem = kmap_atomic(page); |
32ab0a3f | 75 | phys_addr_t pmem_off = sector * 512 + pmem->data_offset; |
61031952 | 76 | void __pmem *pmem_addr = pmem->virt_addr + pmem_off; |
9e853f23 | 77 | |
59e64739 DW |
78 | if (unlikely(is_bad_pmem(&pmem->bb, sector, len))) |
79 | bad_pmem = true; | |
80 | ||
9e853f23 | 81 | if (rw == READ) { |
59e64739 | 82 | if (unlikely(bad_pmem)) |
b5ebc8ec DW |
83 | rc = -EIO; |
84 | else { | |
fc0c2028 | 85 | rc = memcpy_from_pmem(mem + off, pmem_addr, len); |
b5ebc8ec DW |
86 | flush_dcache_page(page); |
87 | } | |
9e853f23 | 88 | } else { |
0a370d26 DW |
89 | /* |
90 | * Note that we write the data both before and after | |
91 | * clearing poison. The write before clear poison | |
92 | * handles situations where the latest written data is | |
93 | * preserved and the clear poison operation simply marks | |
94 | * the address range as valid without changing the data. | |
95 | * In this case application software can assume that an | |
96 | * interrupted write will either return the new good | |
97 | * data or an error. | |
98 | * | |
99 | * However, if pmem_clear_poison() leaves the data in an | |
100 | * indeterminate state we need to perform the write | |
101 | * after clear poison. | |
102 | */ | |
9e853f23 | 103 | flush_dcache_page(page); |
61031952 | 104 | memcpy_to_pmem(pmem_addr, mem + off, len); |
59e64739 DW |
105 | if (unlikely(bad_pmem)) { |
106 | pmem_clear_poison(pmem, pmem_off, len); | |
107 | memcpy_to_pmem(pmem_addr, mem + off, len); | |
108 | } | |
9e853f23 RZ |
109 | } |
110 | ||
111 | kunmap_atomic(mem); | |
b5ebc8ec | 112 | return rc; |
9e853f23 RZ |
113 | } |
114 | ||
dece1635 | 115 | static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio) |
9e853f23 | 116 | { |
e10624f8 | 117 | int rc = 0; |
f0dc089c DW |
118 | bool do_acct; |
119 | unsigned long start; | |
9e853f23 | 120 | struct bio_vec bvec; |
9e853f23 | 121 | struct bvec_iter iter; |
bd842b8c | 122 | struct pmem_device *pmem = q->queuedata; |
9e853f23 | 123 | |
f0dc089c | 124 | do_acct = nd_iostat_start(bio, &start); |
e10624f8 DW |
125 | bio_for_each_segment(bvec, bio, iter) { |
126 | rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, | |
127 | bvec.bv_offset, bio_data_dir(bio), | |
128 | iter.bi_sector); | |
129 | if (rc) { | |
130 | bio->bi_error = rc; | |
131 | break; | |
132 | } | |
133 | } | |
f0dc089c DW |
134 | if (do_acct) |
135 | nd_iostat_end(bio, start); | |
61031952 RZ |
136 | |
137 | if (bio_data_dir(bio)) | |
138 | wmb_pmem(); | |
139 | ||
4246a0b6 | 140 | bio_endio(bio); |
dece1635 | 141 | return BLK_QC_T_NONE; |
9e853f23 RZ |
142 | } |
143 | ||
144 | static int pmem_rw_page(struct block_device *bdev, sector_t sector, | |
145 | struct page *page, int rw) | |
146 | { | |
bd842b8c | 147 | struct pmem_device *pmem = bdev->bd_queue->queuedata; |
e10624f8 | 148 | int rc; |
9e853f23 | 149 | |
09cbfeaf | 150 | rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, rw, sector); |
ba8fe0f8 RZ |
151 | if (rw & WRITE) |
152 | wmb_pmem(); | |
9e853f23 | 153 | |
e10624f8 DW |
154 | /* |
155 | * The ->rw_page interface is subtle and tricky. The core | |
156 | * retries on any error, so we can only invoke page_endio() in | |
157 | * the successful completion case. Otherwise, we'll see crashes | |
158 | * caused by double completion. | |
159 | */ | |
160 | if (rc == 0) | |
161 | page_endio(page, rw & WRITE, 0); | |
162 | ||
163 | return rc; | |
9e853f23 RZ |
164 | } |
165 | ||
166 | static long pmem_direct_access(struct block_device *bdev, sector_t sector, | |
34c0fd54 | 167 | void __pmem **kaddr, pfn_t *pfn) |
9e853f23 | 168 | { |
bd842b8c | 169 | struct pmem_device *pmem = bdev->bd_queue->queuedata; |
32ab0a3f | 170 | resource_size_t offset = sector * 512 + pmem->data_offset; |
589e75d1 | 171 | |
e2e05394 | 172 | *kaddr = pmem->virt_addr + offset; |
34c0fd54 | 173 | *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags); |
9e853f23 | 174 | |
cfe30b87 | 175 | return pmem->size - pmem->pfn_pad - offset; |
9e853f23 RZ |
176 | } |
177 | ||
178 | static const struct block_device_operations pmem_fops = { | |
179 | .owner = THIS_MODULE, | |
180 | .rw_page = pmem_rw_page, | |
181 | .direct_access = pmem_direct_access, | |
58138820 | 182 | .revalidate_disk = nvdimm_revalidate_disk, |
9e853f23 RZ |
183 | }; |
184 | ||
030b99e3 DW |
185 | static void pmem_release_queue(void *q) |
186 | { | |
187 | blk_cleanup_queue(q); | |
188 | } | |
189 | ||
190 | void pmem_release_disk(void *disk) | |
191 | { | |
192 | del_gendisk(disk); | |
193 | put_disk(disk); | |
194 | } | |
195 | ||
200c79da DW |
196 | static int pmem_attach_disk(struct device *dev, |
197 | struct nd_namespace_common *ndns) | |
9e853f23 | 198 | { |
200c79da DW |
199 | struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev); |
200 | struct vmem_altmap __altmap, *altmap = NULL; | |
201 | struct resource *res = &nsio->res; | |
202 | struct nd_pfn *nd_pfn = NULL; | |
203 | int nid = dev_to_node(dev); | |
204 | struct nd_pfn_sb *pfn_sb; | |
9e853f23 | 205 | struct pmem_device *pmem; |
200c79da | 206 | struct resource pfn_res; |
468ded03 | 207 | struct request_queue *q; |
200c79da DW |
208 | struct gendisk *disk; |
209 | void *addr; | |
210 | ||
211 | /* while nsio_rw_bytes is active, parse a pfn info block if present */ | |
212 | if (is_nd_pfn(dev)) { | |
213 | nd_pfn = to_nd_pfn(dev); | |
214 | altmap = nvdimm_setup_pfn(nd_pfn, &pfn_res, &__altmap); | |
215 | if (IS_ERR(altmap)) | |
216 | return PTR_ERR(altmap); | |
217 | } | |
218 | ||
219 | /* we're attaching a block device, disable raw namespace access */ | |
220 | devm_nsio_disable(dev, nsio); | |
9e853f23 | 221 | |
708ab62b | 222 | pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL); |
9e853f23 | 223 | if (!pmem) |
200c79da | 224 | return -ENOMEM; |
9e853f23 | 225 | |
200c79da | 226 | dev_set_drvdata(dev, pmem); |
9e853f23 RZ |
227 | pmem->phys_addr = res->start; |
228 | pmem->size = resource_size(res); | |
96601adb | 229 | if (!arch_has_wmb_pmem()) |
61031952 | 230 | dev_warn(dev, "unable to guarantee persistence of writes\n"); |
9e853f23 | 231 | |
947df02d DW |
232 | if (!devm_request_mem_region(dev, res->start, resource_size(res), |
233 | dev_name(dev))) { | |
234 | dev_warn(dev, "could not reserve region %pR\n", res); | |
200c79da | 235 | return -EBUSY; |
9e853f23 RZ |
236 | } |
237 | ||
468ded03 DW |
238 | q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev)); |
239 | if (!q) | |
200c79da | 240 | return -ENOMEM; |
468ded03 | 241 | |
34c0fd54 | 242 | pmem->pfn_flags = PFN_DEV; |
200c79da DW |
243 | if (is_nd_pfn(dev)) { |
244 | addr = devm_memremap_pages(dev, &pfn_res, &q->q_usage_counter, | |
245 | altmap); | |
246 | pfn_sb = nd_pfn->pfn_sb; | |
247 | pmem->data_offset = le64_to_cpu(pfn_sb->dataoff); | |
248 | pmem->pfn_pad = resource_size(res) - resource_size(&pfn_res); | |
249 | pmem->pfn_flags |= PFN_MAP; | |
250 | res = &pfn_res; /* for badblocks populate */ | |
251 | res->start += pmem->data_offset; | |
252 | } else if (pmem_should_map_pages(dev)) { | |
253 | addr = devm_memremap_pages(dev, &nsio->res, | |
5c2c2587 | 254 | &q->q_usage_counter, NULL); |
34c0fd54 DW |
255 | pmem->pfn_flags |= PFN_MAP; |
256 | } else | |
200c79da DW |
257 | addr = devm_memremap(dev, pmem->phys_addr, |
258 | pmem->size, ARCH_MEMREMAP_PMEM); | |
b36f4761 | 259 | |
030b99e3 DW |
260 | /* |
261 | * At release time the queue must be dead before | |
262 | * devm_memremap_pages is unwound | |
263 | */ | |
264 | if (devm_add_action(dev, pmem_release_queue, q)) { | |
468ded03 | 265 | blk_cleanup_queue(q); |
200c79da | 266 | return -ENOMEM; |
468ded03 | 267 | } |
8c2f7e86 | 268 | |
200c79da DW |
269 | if (IS_ERR(addr)) |
270 | return PTR_ERR(addr); | |
271 | pmem->virt_addr = (void __pmem *) addr; | |
9e853f23 | 272 | |
5a92289f DW |
273 | blk_queue_make_request(q, pmem_make_request); |
274 | blk_queue_physical_block_size(q, PAGE_SIZE); | |
275 | blk_queue_max_hw_sectors(q, UINT_MAX); | |
276 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); | |
277 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); | |
278 | q->queuedata = pmem; | |
9e853f23 | 279 | |
538ea4aa | 280 | disk = alloc_disk_node(0, nid); |
030b99e3 DW |
281 | if (!disk) |
282 | return -ENOMEM; | |
283 | if (devm_add_action(dev, pmem_release_disk, disk)) { | |
284 | put_disk(disk); | |
8c2f7e86 DW |
285 | return -ENOMEM; |
286 | } | |
9e853f23 | 287 | |
9e853f23 | 288 | disk->fops = &pmem_fops; |
5a92289f | 289 | disk->queue = q; |
9e853f23 | 290 | disk->flags = GENHD_FL_EXT_DEVT; |
5212e11f | 291 | nvdimm_namespace_disk_name(ndns, disk->disk_name); |
32ab0a3f | 292 | disk->driverfs_dev = dev; |
cfe30b87 DW |
293 | set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset) |
294 | / 512); | |
b95f5f43 DW |
295 | if (devm_init_badblocks(dev, &pmem->bb)) |
296 | return -ENOMEM; | |
200c79da | 297 | nvdimm_badblocks_populate(to_nd_region(dev->parent), &pmem->bb, res); |
57f7f317 | 298 | disk->bb = &pmem->bb; |
9e853f23 | 299 | add_disk(disk); |
58138820 | 300 | revalidate_disk(disk); |
9e853f23 | 301 | |
8c2f7e86 DW |
302 | return 0; |
303 | } | |
9e853f23 | 304 | |
9f53f9fa | 305 | static int nd_pmem_probe(struct device *dev) |
9e853f23 | 306 | { |
8c2f7e86 | 307 | struct nd_namespace_common *ndns; |
9e853f23 | 308 | |
8c2f7e86 DW |
309 | ndns = nvdimm_namespace_common_probe(dev); |
310 | if (IS_ERR(ndns)) | |
311 | return PTR_ERR(ndns); | |
bf9bccc1 | 312 | |
200c79da DW |
313 | if (devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev))) |
314 | return -ENXIO; | |
708ab62b | 315 | |
200c79da | 316 | if (is_nd_btt(dev)) |
708ab62b CH |
317 | return nvdimm_namespace_attach_btt(ndns); |
318 | ||
32ab0a3f | 319 | if (is_nd_pfn(dev)) |
200c79da | 320 | return pmem_attach_disk(dev, ndns); |
32ab0a3f | 321 | |
200c79da DW |
322 | /* if we find a valid info-block we'll come back as that personality */ |
323 | if (nd_btt_probe(dev, ndns) == 0 || nd_pfn_probe(dev, ndns) == 0) | |
32ab0a3f | 324 | return -ENXIO; |
32ab0a3f | 325 | |
200c79da DW |
326 | /* ...otherwise we're just a raw pmem device */ |
327 | return pmem_attach_disk(dev, ndns); | |
9e853f23 RZ |
328 | } |
329 | ||
9f53f9fa | 330 | static int nd_pmem_remove(struct device *dev) |
9e853f23 | 331 | { |
8c2f7e86 | 332 | if (is_nd_btt(dev)) |
298f2bc5 | 333 | nvdimm_namespace_detach_btt(to_nd_btt(dev)); |
9e853f23 RZ |
334 | return 0; |
335 | } | |
336 | ||
71999466 DW |
337 | static void nd_pmem_notify(struct device *dev, enum nvdimm_event event) |
338 | { | |
a3901802 | 339 | struct nd_region *nd_region = to_nd_region(dev->parent); |
298f2bc5 DW |
340 | struct pmem_device *pmem = dev_get_drvdata(dev); |
341 | resource_size_t offset = 0, end_trunc = 0; | |
342 | struct nd_namespace_common *ndns; | |
343 | struct nd_namespace_io *nsio; | |
344 | struct resource res; | |
71999466 DW |
345 | |
346 | if (event != NVDIMM_REVALIDATE_POISON) | |
347 | return; | |
348 | ||
298f2bc5 DW |
349 | if (is_nd_btt(dev)) { |
350 | struct nd_btt *nd_btt = to_nd_btt(dev); | |
351 | ||
352 | ndns = nd_btt->ndns; | |
353 | } else if (is_nd_pfn(dev)) { | |
a3901802 DW |
354 | struct nd_pfn *nd_pfn = to_nd_pfn(dev); |
355 | struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb; | |
356 | ||
298f2bc5 DW |
357 | ndns = nd_pfn->ndns; |
358 | offset = pmem->data_offset + __le32_to_cpu(pfn_sb->start_pad); | |
359 | end_trunc = __le32_to_cpu(pfn_sb->end_trunc); | |
360 | } else | |
361 | ndns = to_ndns(dev); | |
a3901802 | 362 | |
298f2bc5 DW |
363 | nsio = to_nd_namespace_io(&ndns->dev); |
364 | res.start = nsio->res.start + offset; | |
365 | res.end = nsio->res.end - end_trunc; | |
a3901802 | 366 | nvdimm_badblocks_populate(nd_region, &pmem->bb, &res); |
71999466 DW |
367 | } |
368 | ||
9f53f9fa DW |
369 | MODULE_ALIAS("pmem"); |
370 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO); | |
bf9bccc1 | 371 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM); |
9f53f9fa DW |
372 | static struct nd_device_driver nd_pmem_driver = { |
373 | .probe = nd_pmem_probe, | |
374 | .remove = nd_pmem_remove, | |
71999466 | 375 | .notify = nd_pmem_notify, |
9f53f9fa DW |
376 | .drv = { |
377 | .name = "nd_pmem", | |
9e853f23 | 378 | }, |
bf9bccc1 | 379 | .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM, |
9e853f23 RZ |
380 | }; |
381 | ||
382 | static int __init pmem_init(void) | |
383 | { | |
55155291 | 384 | return nd_driver_register(&nd_pmem_driver); |
9e853f23 RZ |
385 | } |
386 | module_init(pmem_init); | |
387 | ||
388 | static void pmem_exit(void) | |
389 | { | |
9f53f9fa | 390 | driver_unregister(&nd_pmem_driver.drv); |
9e853f23 RZ |
391 | } |
392 | module_exit(pmem_exit); | |
393 | ||
394 | MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>"); | |
395 | MODULE_LICENSE("GPL v2"); |